Selling email archives for AI training

For most companies that have been around more than a few years, email is the longest continuous record of how the business works. Before Slack, before the CRM, before the project management tool, there was email, and it's still where customers, vendors, partners and outside advisors interact with you.

That makes an email archive potentially one of the larger and more valuable datasets you hold. It's also the one that involves the most people outside your company. This guide covers what's valuable, how to scope it, and how to handle it responsibly.

Why email is valuable to AI labs

• External coordination: quotes, orders, negotiations, scheduling and problem resolution with customers and vendors
• Formal decision records: approvals, sign-offs and instructions that were put in writing on purpose
• Long time depth: often a decade or more of history
• Document context: attachments such as proposals, contract drafts and reports, with the discussion around them
• Thread structure: clear reply chains showing how a conversation develops

Compared with chat, email is more deliberate and more external. Together, the two show how internal discussion turns into external action.

Which mailboxes and folders matter most

• Operational roles: operations, purchasing, customer service, project management, account management, dispatch, estimating
• Shared mailboxes: orders@, support@, ap@, quotes@ and similar, which are often highly structured and process-driven
• Long-tenured users whose mailboxes span many years

Lower-value content includes newsletters, marketing emails received, notifications, automated reports, calendar invitations and spam.

What to exclude

Agree on exclusions before any export. Common ones:

• Executive, HR, legal and finance leadership mailboxes, or at least the sensitive folders within them
• Privileged communications with outside counsel
• Board, investor and M&A correspondence
• Personal email that employees sent from work accounts
• Correspondence with customers whose contracts restrict use of their communications
• Regulated information: health information, payment card data, Social Security numbers, financial account details
• Mail under legal hold, which must be preserved, and which you may want to keep out of any third-party project entirely
• Attachments in high-risk categories (scanned IDs, signed contracts, medical or financial documents)

The outside-party question

Unlike Slack, much of your email was written by people who don't work for you: customers, vendors, applicants, advisors. They agreed to nothing with respect to your data programs. That doesn't necessarily prevent licensing, but it raises the bar:

• De-identification needs to cover outside parties as thoroughly as employees, including names, companies, email domains and signatures.
• Your privacy policy and customer agreements may limit how you use communications from customers.
• Some state privacy laws regulate selling or sharing personal information, with different treatment for properly de-identified data.

This isn't legal advice. Have counsel review the proposed scope, your privacy commitments and any customer contracts before licensing email data.

How email exports work

Your options depend on your platform, plan and admin permissions, so confirm current details with your vendor or IT admin. In general:

• Google Workspace administrators can export organization data or use Google Vault (on editions that include it) to search and export mail.
• Microsoft 365 administrators can use Microsoft Purview eDiscovery or content search to export mailboxes, typically as PST files, depending on licensing.
• On-premises Exchange or archive tools (Mimecast, Proofpoint and others) usually have their own export methods.

Exports are large. A partner's engineers can usually handle extraction, parsing and cleanup, with your IT admin granting scoped access and your team approving which mailboxes and date ranges are included.

De-identifying email

Email is among the harder data types to de-identify because identifiers appear everywhere:

1. Headers: sender, recipients, CC lists and domains. Replace them with consistent pseudonyms ("Person 22 at Vendor 9").
2. Signatures: names, titles, phone numbers, addresses and legal disclaimers
3. Quoted replies and forwards, which repeat earlier content, including identifiers
4. Body text: names, account numbers, order numbers, addresses and anything pasted in
5. Attachments: file names, document metadata and contents, which often need separate processing or exclusion
6. Embedded images such as logos and screenshots

Consistent pseudonyms matter: the same vendor contact should be "Person 22 at Vendor 9" in every thread so the relationship stays readable. Review samples by hand. See our de-identification guide.

What affects the value

• Years of history and how continuous it is
• Share of substantive human correspondence versus automated mail
• Operational breadth: many roles and functions represented
• Attachments included (after processing), which add document context
• Connection to other systems: email alongside Slack, CRM or ticketing shows complete workflows
• Exclusions: excluding the most active operational mailboxes reduces value significantly

Offers vary with these factors. See how much is my company's data worth?

A practical checklist

• Confirm your email platform, edition and what export tools your admin can use.
• Pull rough numbers: mailbox count, total storage, oldest messages.
• List mailboxes and shared mailboxes as include, exclude or review.
• Check for active legal holds and retention policies.
• Review your privacy policy, acceptable-use policy and key customer contracts.
• Decide who will run extraction and de-identification, and how you'll review a sample.

Getting started

To get an offer, share rough estimates: your email platform, roughly how many mailboxes and how much storage, years of history, which other systems you might include, headcount and years in business. No mail needs to be exported to get an offer.

DataOffer packages and de-identifies email data (or supports your team in doing so), gets competing offers from multiple labs, and brings you the best one. You choose which mailboxes are included, and nothing is shared until you approve the buyer, price and terms.

This guide is general information, not legal, tax or financial advice. Figures and ranges are illustrative; talk to qualified advisors about your situation.